Hi, I'm Liam

I'm a senior EECS student at UC Berkeley, and I'm interested in ML! I'm actively seeking internship opportunities in AI/ML. I'll be showcasing my personal projects here.

Liam Newsam

Education

University of California, Berkeley Berkeley, CA
Bachelor of Electrical Engineering & Computer Science Aug. 2023 – Present
  • GPA: 4.0/4.0
  • Selected Coursework: Algorithms, Data Structures, Machine Learning, Operating Systems, Reinforcement Learning

Experience

Undergraduate Research Assistant Jan. 2025 – Sep. 2026
University of California, Berkeley | Dr. Taubinsky, Dr. Shenhav Berkeley, CA
  • Working on the Social Signals research project which is investigating job discrimination in roommate requests.
  • Developed a GitHub Workflow automated system written in Python which scrapes web data, sends and receives emails, and uses Pandas to process large amounts of online data.
  • Designed a GitHub Pages website using HTML and CSS for participants of the study to review their contributions.
Network Manager Jan. 2025 – Sep. 2026
Kingman Student Cooperative Berkeley, CA
  • Maintaining network infrastructure for a 50-person residence by optimizing the configurations of 13 access points across 4 floors, ensuring sufficient power draws, and debugging network switches and cable connections. I received excellent reviews from many house members.
Undergraduate Researcher June 2024 – Aug. 2024
Temple University HCI Lab | Dr. MacNeil, Cindy Zastudil Philadelphia, PA
  • Participated in the NSF-funded REU Site in Pervasive Computing for Smart Health, Safety, and Well-being at the Human Computer Interface Lab.
  • Worked on a project to improve communication options for verbally impaired autistic children.
  • Designed and created a complete React front-end prototype for a novel Visual Screen Display which uses LLMs to automatically generate communication options from on-demand camera photos.
    • Decreased average time-to-create by 13.5 seconds compared to state-of-the-art VSD software (Snap Scene).1
    • Increased median number of finalized hotspots generated from 3 to 5 hotspots per scene.1
  • 1Helping or Homogenizing? GenAI as a Design Partner to Pre-Service SLPs...

Projects

Wizard101 Demonstration | PyTorch, TorchRL, Flask, React, SQLAlchemy, Google Cloud Nov. 2025 – Jun. 2026
  • Created a web demo of the video game Wizard101 with multiplayer capability.
  • Trained a competitive agent using a transformer architecture with PPO and evolution. Trained agents on average win 11 percentage points more than agents that take random actions, demonstrating that learning occurred.
Humanoid RL Training | MuJoCo, PyTorch, TorchRL, Robotics, Physical AI May. 2026 – Sep. 2026
  • Trained a Unitree G1 model in the MuJoCo physics simulator according to the general guidelines described by the paper, "Reinforcement Learning for Versatile, Dynamic, and Robust Bipedal Locomotion Control" (Levine et al.).
  • Trained agent could successfully stand and resist significant wind speeds up to 15 m/s.
  • Trained agent could walk several meters, training is still in progress.
Barebones RISC-V CPU | Logisim, RISC-V Oct. 2024 – Nov. 2024
  • Implemented the 5-stage processing paradigm in the circuit-design application Logisim.
  • Manually-constructed a CPU that recognizes over 30 RISC-V assembly instructions.
  • Enhanced performance through stage piping and parallelized instructions.

Technical Skills